A non-contact image-to-patient registration method using kinect sensor and WAP-ICP

In many medical image-guided navigation systems (IGNS), image-to-patient registration plays an important part for applying reliable anatomical information mapping and spatial guidance. In this study, we propose a totally non-contact image-to-patient registration technique using kinect sensor and an ICP-based (Iterative Closest Point-based) registration algorithm which is named WAP-ICP. A Kinect sensor is used to detect facial feature points form a patient and calculate 3D coordinates of these points. The WAP-ICP algorithm can help us to register these 3D points to the surface reconstructed from the patient's preoperative CT images without pre-alignment. Moreover, WAP-ICP altgorithm uses not only a random-perturbation technique to deal with the local minimum problem of ICP, but also a weighting strategy to reject noisy feature points. Experimental results reveal that the proposed WAP-ICP algorithm has great improvement in robustness than the ICP algorithm.